A pilot study was conducted in 6 public bars in Edinburgh [Scotland, United Kingdom] on 13 weekdays. During 26 sessions of 2.5 h 3 pairs of trained observers watched a total of 1398 people. One observer pair independently counted all alcoholic beverages sold; the 2nd pair counted all those entering the bar to buy or consume drinks; the 3rd pair recorded all drinks consumed by a sample of drinkers who could be unobtrusively but clearly observed. The average interobserver agreement about the amount of beer sold was 96% and about distilled spirits 100%. Agreement was lowest when the total sales were low. Agreement on the number of customers/session also averaged 96%, irrespective of the numbers involved or whether the enumerators had additional tasks. Interobserver agreement on average consumption/drinker averaged 96% for beer and 90% for spirits. Disagreement increased as the amount consumed by the drinker increased. Most disagreement was about amount consumed, not about type of drink. Comparison of the consumption of the study sample with that of the total population of patrons indicated that the sample mean overestimated beer consumption by 13% and underestimated spirits consumption by 1.5%. Individual sessions often showed wider disparities. If used across a series of occasions, the observational technique employed was unlikely to reflect a systematic bias, but findings from a single session should be used cautiously. Direct observational techniques may be best employed in conjunction with surveys of small communities where much of the drinking is done in public.